Why is the score gap so big for quad core mobile phones| Love to play computer games - 上海419贵族宝贝-上海后花园1314-上海品茶网-阿拉爱上海

Recently, there was a piece of news on the Internet, that is, the famous Android scoring software, Anthare, claimed that Huawei's glory four cores were in the Score running software "Cheating". This matter has also attracted wide attention in the industry. It is reported that Huawei Glory 2 "opened its fire" in the test for the score running of rabbit, leading to 2000 points higher in the official version of rabbit 3.0 than in the test version. The four core of glory scores 15000 points, while some mobile phones can even exceed 20000 points. Why?

First, let's take a look at the "cheating" problem claimed by Anthare. Anthare analyzed a system file classs.dex in the Glory quad core, and found that Glory 2 will adjust the frequency of GPU according to some specific "keywords". For example, when encountering the words benchmark, geekbench, vellamo, etc., the highest frequency will be turned on, The GPU runs to 480MHz, but it may only run at 240MHz in some normal situations. In the author's opinion, this is totally unrelated to cheating. When running the run score software to maximize its performance, it usually runs at a reduced frequency due to power consumption and temperature. This seems different from the concept of "deliberate optimization", which can only be said to be a problem in frequency control. Moreover, such a problem should not be raised by a third-party testing software company, and the headline should also be used to indicate that it is "cooking". It can only be said that Anthare has deviated from the objective. Running score software has high frequency and other software has low frequency. This problem will affect the experience of some large games or applications. Shouldn't users complain?

Back to the main topic, the same is the four cores. Why is the difference so big in the score run of Anthare? The following is the author's test of the four core run of glory.

This is the score of Tegra3 and Exynos 4412 for comparison. They all use the kernel based on Cortex-A9, so they are very comparable. In order to make the comparison result more intuitive, the author locked the other two models on the same 1.4G as the Glory quad core. Looking at the total score, the gap between the three is not big (Tegra3 on the left and Exynos 4412 on the right).

Let's take a look at the memory. Because Tegra3 (AP37) uses 1G LPDDR2 32bit memory, its score is lower than Exynos 4412 and K3V2 64bit memory.

There is nothing to say about the performance of integer operations here. The gap generated here really depends on the level of A9 architecture utilization and optimization.

The most interesting GPU part is also the "cheating" of Anthare's attack on the Glory Quad. From the point of view of the score of graphics processing, the integration of the Glory Quad before and after has increased by about 20%, which seems to be really the increase of frequency. As for the comparison of the three, there is a certain gap in the score of graphics processing. The Exynos 4412 basically continues its strong performance in practical applications, surpassing the other two SoCs. However, it is strange that the GC4000 theoretical GPU computing power of K3V2 is 34.6GFLOPS, which is even half higher than the Mali-400MP4 used by Exynos 4412, but its score is worse than that of its competitors. From the 3D run of Anthare, it seems that the run of these three GPUs is a bit "weird". The gap between Exynos 4412 and Tegra3 is not big, and the K3V2 is too low. Even after the test, it was found that the Mali-400MP4 from 350MHz to 800MHz, the 3D running scores of Anthare were almost the same... Personally, I think the GPU was tested using a more professional GLBenchmark2.5.

It can be seen from the architecture and theory that we can analyze many problems, and these differences are also directly reflected in the score of the scoring software. In many cases, the CPU frequency has a greater impact on the score than the difference between different processors. Does that mean that Anthare is very referential?

Finally, APQ8064 scored 21589 points, Since APQ8064 cannot be locked in the same frequency 1.4G comparison through software, it can only use the default 1.5G score, so pay special attention to certain changes in frequency. APQ8064 scores are exaggerated, leaving three A9 architecture processors behind. The scores of memory, floating point operation and 3D graphics processing are much higher than those of other SoCs. Of course, this is also related to the characteristics of APQ8064 itself. The floating point operation performance of the krait of APQ8064 is 3.3DMIPS, while the single core of the A9 architecture is 2.5DMIPS. From the perspective of architecture, krait is indeed superior in floating point operation performance. In terms of its 3D performance, the APQ8064's Adreno 320 is indeed a very powerful GPU. Its computing power has reached 57GFLOPS, and its theoretical performance is even more than twice that of some competitors. Therefore, it is not surprising that it can achieve such dazzling results in its run points.

We can see from the score score of these products that there are many problems that conform to the actual hardware performance and architecture characteristics, but there are still many tricks in the test software, such as the super high score of APQ8064, which seems to be inconsistent with the actual use, Is it really possible to leave the other four cores by nearly 1/4 in running speed in actual use? This is obviously unrealistic. In some actual tests of CPU general performance, it is even inferior to A9 architecture processors in many cases. The super high GPU score is quite consistent with the hardware performance, but in practical applications, some large games also involve the optimization of GPU, so we can not just draw conclusions by looking at the score. As for Anthare's running score software, I advise you to keep it as a reference only and never take it seriously.

Appendix: Reference Table of GPU Computing Capacity

Adreno of APQ8064 320@400MHz :57GFLOPS

Exynos4412 Mali-400MP4@533MHz ：21 GFLOPS

ULP of Terga3 Geforce@520MHz ：12.5 GFLOPS

K3V2 GC4000@480MHz ：34.6 GFLOPS

Learn more about new cool devices, stay tuned